← back

Browser-based Whisper transcription using WebGPU and adaptive model selection

brunochavesj | 2026-02-18 02:23 UTC | source

Free Unlimited Transcription With No Minutes Limit

Cowslator is a free unlimited transcription app for creators, teams, and students who need reliable output without paywalls. You get no minutes limit, plus audio to text free and video to text free workflows from one interface.

The transcription engine is based on Whisper AI. Cowslator supports multiple Whisper models (tiny, base, small, medium, and large, including quantized variants) so users can balance speed and quality by workload.

Need scale? Use batch transcription to upload multiple files or a full folder, then export `.txt`, export SRT, and export LRC. Your work stays private because Cowslator is a private transcription tool built for local browser transcription.

Privacy-first Runs locally in your browser

Open in app Read the unlimited transcription guide

Benefits That Save Time and Protect Your Data

Truly free unlimited transcription

This is free unlimited transcription by design. There is no minutes limit, no trial countdown, and no credit card gate.

Built for real batch workloads

Run batch transcription when you need to upload multiple files. Queue audio and video together without opening each file one by one.

Audio to text free and video to text free

Convert interviews, classes, meetings, and recordings with audio to text free and video to text free output in a single tool.

Subtitle-ready exports

Download clean results in `.txt` plus subtitle formats: export SRT and export LRC for editors, players, and social clips.

Whisper model options for different machines

Use Whisper AI with tiny, base, small, medium, and large model families. Cowslator can filter model choices based on your CPU threads and RAM so the default suggestion matches your machine capability.

Need strategy-level detail? Explore the long-form free unlimited transcription guide.

Privacy Is the Product, Not a Side Note

Cowslator is a private transcription tool created for sensitive workflows. With local browser transcription, your files are processed on your own device and do not have to be uploaded to external servers just to get a transcript.

  • Your media stays on your machine during processing.
  • You decide which files to process and when to delete them.
  • The app supports a privacy-first workflow for legal, healthcare-adjacent, and internal business material.
  • You still get free unlimited transcription and no minutes limit while keeping control.

Why Cowslator vs cloud transcription tools

Decision factor Cowslator Typical cloud transcription tools
Pricing model Free unlimited transcription with no subscription required. Often based on monthly quotas, tiers, or usage credits.
Minutes policy No minutes limit for regular use. Commonly enforces minutes caps or throttling after limits.
Data flow Local browser transcription for privacy-first control. Usually requires file upload to a remote server before processing.
Workload handling Batch transcription and folder queue support. Batch features may be limited by plan tier or per-file billing.
Subtitle output Free `.txt`, export SRT, and export LRC. Advanced export formats may be restricted by paid plans.

Read the full cloud-transcription comparison page for a deeper breakdown.

How It Works in Four Practical Steps

  1. Test your processor. Cowslator filters available Whisper models according to your machine capabilities (CPU threads and RAM) and suggests a safe default.
  2. Choose single upload or batch transcription mode to upload multiple files.
  3. Run audio to text free or video to text free processing with your local resources.
  4. Download results in `.txt`, export SRT, or export LRC.

Customer Feedback (Placeholder)

Use case: Content teams

Placeholder for verified user quote about batch transcription and faster publishing workflows.

Use case: Privacy-sensitive work

Placeholder for verified user quote about using this private transcription tool for internal recordings.

Use case: Students and researchers

Placeholder for verified user quote about free unlimited transcription with no minutes limit.

FAQ

Is Cowslator really free unlimited transcription?

Yes. Cowslator is free unlimited transcription software with no paywall and no forced subscription.

Is there really no minutes limit?

Yes. The product is built around a no minutes limit experience so heavy users are not blocked by quotas.

Can I upload multiple files for batch work?

Yes. Batch transcription supports folder upload so you can upload multiple files in one queue.

Can I do audio to text free and video to text free?

Yes. Both audio to text free and video to text free use cases are supported from the same flow.

What makes this a private transcription tool?

Cowslator is a private transcription tool based on local browser transcription, so files stay under your control on your own machine.

Which AI powers transcription in Cowslator?

Cowslator is powered by Whisper AI. The app supports multiple Whisper model sizes, including tiny, base, small, medium, and large tiers with quantized options.

How does model filtering work?

Model options are filtered by your machine capability. Cowslator uses detected CPU threads and selected RAM to recommend Whisper models that are more likely to run smoothly.

1 points | 1 comments | original link

Comments

brunochavesj | 2026-02-18 02:23 UTC
Hey HN folks,

I built a browser-based audio/video transcription tool that runs Whisper locally using WebGPU (with a CPU fallback via whisper.cpp).

There’s no backend processing — everything runs on the user’s machine. Files are never uploaded to a server.

Some implementation details:

Uses ONNX + WebGPU when available

Falls back to whisper.cpp (WASM) on CPU

Adaptive model selection engine chooses tiny/base/small/medium/etc based on:

CPU cores/threads

RAM (user input)

WebGPU availability

Models are downloaded once and stored in IndexedDB

Batch folder support (you can select a folder with hundreds or thousands of files)

The idea is that scaling happens on the client. A better GPU allows larger models. On weaker machines it automatically falls back.

I’m currently stress-testing it by transcribing 1000 audio files in a single batch to see how stable the memory behavior is over time.

Outputs:

.txt

.srt

.vtt

.lrc

Considering adding a final ZIP export so users don’t have to download files individually.

Would appreciate feedback, especially on:

WebGPU reliability across browsers

Memory management strategies for long batch runs

UX for hardware-based model selection

From what I tested, it seems to work best in chrome, didn't test it with chromium yet.

Curious what you folks think.